The reference-management market acquired an AIera second tier between 2023 and 2026. Where the 2018–2022 comparative literature could treat Zotero, Mendeley, EndNote, and RefWorks as a stable four-corner taxonomy, the 2026 landscape adds AI-augmented entrants — Paperguide, Logically, Paperpile (post-2024 BeyondPDF reposition), and adjacent AI-research tools (Elicit, Consensus, SciSpace) that consume reference-manager territory at the upstream-search boundary. The Document Engineering textbook series Vol 4 (Citation and Reference Management), aimed at Vietnamese learners aged 14 and up, anchors its baseline on Zotero (per node H5.1) but must teach learners how to evaluate the AI-era alternatives without overreacting to the feature-churn cycle. This paper supplies the evaluation framework. Three competitors are selected for the first-edition comparison — Paperguide, Logically, Paperpile — under an explicit cap rationale. Each tool is scored against Zotero across nine dimensions: feature set, AI-summary quality, citation provenance discipline, export fidelity, pricing tier, longevity risk, classroom-deployability, Vietnamese-language fit, and learnercost. Each dimension carries a rating, an evidence trail anchored in vendor documentation or comparative literature, and explicit caveats. The paper supplies a migration-criteria block per competitor (when to leave Zotero, when to stay) and an AIfeature-by-tool table that maps each named AI feature (summarisation, extraction, question-over-library, citation-suggestion) to its provenance-discipline implications. The discussion maps the matrix into Vol 4 future-looking chapters, defends the three-competitor cap, names the limits (feature churn, pricing volatility, vendor longevity), flags five open questions, and proposes an annual refresh cadence with audit-loop ratification. Empirical content is restricted to document analysis and vendordocumented behaviour; no learner trial. Target is the audit-loop rigor bar, not venue submission.
That Le (2026) studied this question.